This is related to my recent thread,

https://list.orgmode.org/87wmi9etku.fsf@gmail.com/T/#m93bd964243638ee358d4375c9ed4f40e066238d4

which addresses problems with prompt filtering of R and python async evaluation in Org 9.7.

Here, I address the non-async case.

First, I introduce options to `org-babel-comint-with-output' to skip the
cleanup of prompts and dangling text.  This allows individual Babel
languages to handle these tasks in a more robust, language-specific
manner, e.g. by sourcing temp files or relying on functionality from
external libraries like python.el or ESS.

In particular, as I discussed in the link above, it is very difficult to
do the prompt cleanup in a way that is robust to false positives.  ob-shell
accomplishes this by changing the prompt to a unique string, but this is
not an option for other languages, and has its own drawbacks (e.g. with
conda).  ob-python takes a different approach, and evaluates code in a
way that avoids leaking prompts, thereby skipping the need to clean them
up.

Previously, ob-python had eschewed using `org-babel-comint-with-output',
and re-implemented its own version without the cleanup steps.  With this
patch, ob-python can go back to using `org-babel-comint-with-output',
thereby reducing code duplication.

Finally, I add a new implementation for R non-async output, that follows
the ob-python approach of avoiding prompt leakage altogether, thereby
skipping the need for a cleanup step. I add a unit test to demonstrate
the improved robustness of this approach: previously, this test would
fail due to a false positive prompt, but it passes with the new
implementation.